*** Update: This became less of a book as I realized how much of a n00b I was and more of a collection of materials to provide intuitions that might one day get compiled into a book.
This book is a WIP. But while working on my own machine learning projects I found myself struggling a bit to understand why certain solutions worked while others didn't. The idea of memory for RNNs and LSTMs confused me and I wasn't clear on how the models played into reinforcement learning algorithms like PPO.
Enter Machine Learning Intuitions
- which is a book I started to write that will introduce various concepts, math and provide toy, but working implementations you can study and tweak to help you understand how these different solutions work and hopefully provide better intuitions about how to employ them in solutions.
While not many of us will need to hand code an LSTM from scratch, understanding how hidden layers function as memory is very important to being able to deploy LSTM in the right places.